Logistics demand prediction method and system based on AI

Abstract

The invention relates to the technical field of logistics management, in particular to an AI-based logistics demand prediction method and system, and the method comprises the following steps: recognizing nodes based on a period difference value and high-frequency fluctuation, constructing a graph link, and predicting a logistics demand trend. According to the method, accurate screening and positioning of key time nodes are realized by introducing inter-node difference accumulated absolute difference and a local high-frequency fluctuation accumulation mechanism and fusing periodic change and high-frequency feature recognition, a path structure is constructed through inter-node time sequence intervals, and a map link relationship is extracted; a logistics demand potential fluctuation rule is extracted from a sequence and a connection rule of a map path, and then is converted into standard structure data through AI for prediction output, so that the problem that a traditional model is insensitive to non-linear fluctuation identification can be effectively solved, the response capability to multiple change trends is enhanced, and the prediction accuracy is improved. The adaptability and effectiveness of the prediction result in a dynamic scene are improved, and the identification precision and prediction granularity of the multi-stage multi-feature logistics demand are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics management, in particular to a logistics demand prediction method and system based on AI. BACKGROUND

[0002] The technical field of logistics management relates to the planning, implementation and control of the entire process from the supply end to the demand end of materials, including transportation scheduling, warehouse control, inventory optimization, distribution path planning and supply chain collaboration. Through dynamic collection, analysis and processing of information flow and material flow at each node in the logistics chain, the efficiency of resource allocation is improved and the cost is controlled. Among them, the logistics demand prediction method refers to the technology of predicting the trend of future logistics demand based on historical logistics data. It is aimed at predicting the quantity and time distribution of future material demand in logistics activities. It usually uses time series analysis method, regression model or rule-based statistical method to analyze historical transportation volume, warehouse entry and exit records, order data and other small information to build a demand trend model to complete the prediction process.

[0003] In the current logistics demand prediction process, the method relying on traditional time series and statistical models is easily limited by linear assumptions and rule settings. It is slow to respond to scenarios with unclear periodicity or high-frequency local changes, especially in nodes with frequent sudden demand fluctuations. It is often difficult to capture key change points in time, which leads to the accumulation of prediction bias and causes scheduling delays and uneven resource allocation. For example, during the peak of concentrated orders before holidays or temporary allocation, the model cannot effectively identify key nodes, resulting in problems such as periodic congestion and service response lag in the material supply chain. SUMMARY

[0004] In order to solve the technical problems existing in the prior art, the embodiments of the present application provide a logistics demand prediction method based on AI, comprising the following steps: In order to achieve the above purpose, the present application adopts the following technical scheme: a logistics demand prediction method based on AI, comprising the following steps: S1: Obtain a sequence of continuous order quantities of materials, read the quantity difference of adjacent order quantities, sequentially execute the cumulative absolute difference judgment and mark the trigger node, generate a sequence of periodic node extraction numbers; S2: According to the sequence of periodic node extraction numbers, obtain the order quantity records of the corresponding time period of material allocation, construct a continuous window according to time sliding, when the cumulative difference value in the window is greater than the local high-frequency fluctuation recognition threshold, mark the window as a demand intensive node window, generate a high-frequency label index matching sequence; S3: According to the periodic node extraction number sequence and the high-frequency label index matching sequence, the intersection nodes with both periodicity and high-frequency labels are selected to generate a periodic high-frequency composite positioning number sequence; S4: Based on the periodic high-frequency composite positioning number sequence, the node sequence is sorted in the time axis, the interval index length of adjacent nodes is recorded, the composite nodes are connected to build a graph path, and a graph node arrangement mapping index is established to generate a graph path link structure number table; S5: According to the graph path link structure number table, the node time order and the path number order are read, the logistics demand fluctuation law in the graph structure is identified by AI, and is converted into a standard format data to generate an AI logistics demand prediction record.

[0005] As a further scheme of the application, the periodic node extraction number sequence includes node number identification, trigger time index, periodic starting position, and periodic fluctuation amplitude, the high-frequency label index matching sequence includes dense node number, high-frequency fluctuation amplitude, window start and end time, and label positioning weight, the periodic high-frequency composite positioning number sequence includes composite node number, node intersection index, periodic and high-frequency attribute identification, and node time order mark, the graph path link structure number table includes path number, node connection relationship, node interval distance, and graph mapping index, and the AI logistics demand prediction record includes prediction time node, demand change trend, and standard structure data format.

[0006] As a further scheme of the application, the periodic node extraction number sequence is obtained by: S111: Obtain a continuous order quantity sequence of goods, and perform difference operation on the order quantities of adjacent two orders in the sequence to generate a difference sequence according to the original order time sequence, and perform sequential aggregation operation on every five continuous difference value data in the difference sequence to generate a continuous difference value sliding set; S112: Based on each group of difference value data in the continuous difference value sliding set, perform summation processing on the absolute values of all difference value data, if the summation value is greater than the periodic quantity fluctuation judgment threshold for the first time, record the time index position of the corresponding difference value group, and generate a fluctuation threshold trigger index value; S113: According to the fluctuation threshold trigger index value, locate the corresponding time point in the difference value sequence, and take the time point as the starting point of the periodic node, sequentially extract all subsequent index positions that satisfy the same judgment condition, arrange all extracted index values in time sequence to generate a periodic node extraction number sequence.

[0007] As a further scheme of the application, the high-frequency label index matching sequence is obtained by: S211: Extract the number sequence according to the periodic node, obtain the material order quantity record within the corresponding time period, and construct the order quantity subsequence of equal length interval before and after the node on the time axis based on the index of each node. Set a sliding window structure of the same length for each subsequence, extract and record the three order quantity values ​​in each window one by one according to the time order, and generate the sliding window order difference unit. S212: Based on the three order quantity data of each window in the sliding window order difference unit, perform continuous difference calculations in sequence and accumulate the results. If the accumulated difference is greater than the local high-frequency fluctuation identification threshold, record the corresponding window as a demand-intensive area window and generate a demand fluctuation exceeding limit index set. S213: Filter windows in the time series according to the index position in the demand fluctuation excess index set, arrange all window indexes that meet the conditions in order, add high-frequency demand tags to each identified window, integrate all marked window indexes and establish a complete index sequence, and generate a high-frequency tag index matching sequence.

[0008] As a further aspect of the present invention, the step of obtaining the periodic high-frequency composite positioning number sequence is as follows: S311: Based on the extracted number sequence of the periodic node and the matching sequence of the high-frequency label index, perform a comparison operation on the time index in the sequence. If both sequences include a certain time index, record the corresponding index as the intersection node index and generate a periodic label intersection index set. S312: Based on the set of periodic label intersection indexes, screen the position of each index value in the original order sequence, exclude nodes that exceed the current scheduling cycle boundary or are in a data missing state, and retain the corresponding index number for the remaining valid nodes to obtain a list of available intersection node indexes. S313: Based on all index numbers in the available intersection node index list, according to the arrangement order on the order data time axis, the numbers are reordered, and each node is assigned a continuous number identifier to generate a periodic high-frequency composite positioning number sequence.

[0009] As a further aspect of the present invention, the comparison operation specifically involves sequentially scanning each time index value in the period number using a step-by-step matching method, and searching for the existence of a matching index in the high-frequency tag sequence.

[0010] As a further aspect of the present invention, the step of obtaining the map path link structure number table is as follows: S411: Based on the periodic high-frequency composite positioning number sequence, perform an ascending sort operation on all node indices in the sequence, rearrange each node according to the chronological order, construct a time index difference sequence between adjacent nodes, extract and number all the differences, and obtain a node interval index distance sequence. S412: Based on the node interval index distance sequence, construct a connection path structure by using each node index and its subsequent node interval index as edge weights, connect the composite nodes in the form of adjacency pairs to construct a path graph, calculate and obtain the total node connection strength, and perform structural organization to obtain node path connection strength data. S413: Based on the connection pair sequence structure corresponding to the node path connection strength data, establish a path graph index mapping, extract the start and end node indexes and connection numbers of each connection segment, and store all path segments sequentially in a structured order to obtain the graph path link structure number table.

[0011] As a further aspect of the present invention, the step of obtaining the AI ​​logistics demand forecast record is as follows: S511: Based on the graph path link structure number table, obtain the corresponding structure of node index and path number, extract the time index of the start and end nodes of each path segment, reorder all path numbers in time order, construct a path number sequence structure that advances in time, perform continuity verification on the path number sequence and mark time features, and establish a graph path time sequence arrangement matrix. S512: Based on the path sequence and time index information in the path time sequence arrangement matrix of the map, monitor the fluctuation pattern of the path structure in the time dimension, perform periodic behavior recognition and extract feature label information according to the path change frequency and structural jump behavior, and generate a logistics path fluctuation feature vector. S513: Based on the structural label information extracted from the logistics path fluctuation feature vector, each label is uniformly encoded and mapped, decomposed into a set of data fields in a standard format, construct a structured data frame, and establish an AI logistics demand prediction record.

[0012] As a further aspect of the present invention, the monitoring path structure fluctuation pattern in the time dimension specifically involves aggregating the length difference, time span difference, and path number increment of continuous paths to determine the fluctuation pattern of the path sequence.

[0013] An AI-based logistics demand forecasting system includes: The periodic difference extraction module is used to execute S1: obtain the sequence of consecutive order quantities of materials, read the quantity of adjacent orders to calculate the quantity difference, perform cumulative absolute difference judgment and mark the trigger node in sequence, and generate a periodic node extraction number sequence; The window label positioning module is used to execute S2: extract the number sequence according to the periodic node, obtain the order quantity record of the material allocation corresponding time period, construct a continuous window by sliding according to time, and when the cumulative difference in the window is greater than the local high frequency fluctuation identification threshold, the window is marked as a demand-intensive node window and a high frequency label index matching sequence is generated. The node intersection identification module is used to perform S3: based on the number sequence extracted from the periodic node and the matching sequence of the high-frequency tag index, select the intersection node that has both periodic attributes and high-frequency tags, and generate a periodic high-frequency composite positioning number sequence; The graph path construction module is used to execute S4: based on the periodic high-frequency composite positioning number sequence, sort the node sequence on the time axis, record the interval index length of adjacent nodes, connect each composite node to construct the graph path, establish a graph node arrangement mapping index, and generate a graph path link structure number table. The prediction data generation module is used to execute S5: based on the number table of the graph path link structure, read the node time sequence and path number sequence, identify the logistics demand fluctuation pattern in the graph structure through AI, and convert it into standard format data to generate AI logistics demand prediction records.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention introduces a mechanism for accumulating the absolute difference between nodes and accumulating local high-frequency fluctuations, integrating periodic changes with high-frequency feature recognition to achieve accurate screening and positioning of key time nodes. By constructing a path structure through temporal intervals between nodes and extracting graph link relationships, the potential fluctuation patterns of logistics demand are extracted from the sequence and connection rules of the graph path. This data is then converted into standard structured data by AI for prediction output. This effectively overcomes the problem of insensitivity to nonlinear fluctuation recognition in traditional models, enhances the response capability to multiple changing trends, improves the adaptability and effectiveness of prediction results in dynamic scenarios, and significantly improves the recognition accuracy and prediction granularity of multi-stage and multi-feature logistics demand. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a flowchart of the process for obtaining the number sequence of the periodic nodes in this invention; Figure 3This is a flowchart of the high-frequency tag index matching sequence acquisition process of the present invention; Figure 4 This is a flowchart of the process for obtaining the periodic high-frequency composite positioning number sequence of the present invention; Figure 5 This is a flowchart of the process for obtaining the map path link structure number table of the present invention; Figure 6 This is a flowchart of the AI ​​logistics demand prediction record acquisition process of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides an AI-based logistics demand forecasting method, comprising the following steps: S1: Obtain the sequence of consecutive order quantities of materials, read the quantity of adjacent orders to calculate the quantity difference, and perform cumulative absolute difference judgment on every five difference contents in turn. When the cumulative result reaches the periodic quantity fluctuation judgment threshold for the first time, mark the current time index as the trigger node and generate a periodic node extraction number sequence. S2: Extract the number sequence based on the periodic nodes, obtain the order quantity records of the corresponding time period for material allocation, construct a continuous window by sliding it over time, and accumulate the difference between three consecutive orders in each sliding window. When the accumulated difference in the window is greater than the local high-frequency fluctuation identification threshold, the window is marked as a demand-intensive node window, and a high-frequency label index matching sequence is generated. S3: Based on the periodic node extracted number sequence and the high-frequency tag index matching sequence, compare the two sequences, perform screening processing on the nodes in the matching time index, select the intersection node that has both periodic attributes and high-frequency tags, and record its time distribution order to generate a periodic high-frequency composite positioning number sequence. S4: Based on the periodic high-frequency composite positioning number sequence, the node sequence is sorted on the time axis, the interval index length of adjacent nodes is recorded, the composite nodes are connected to construct the graph path, and a graph node arrangement mapping index is established to generate a graph path link structure number table. S5: Based on the numbering table of the path link structure in the graph, read the node time sequence and path number sequence, identify the logistics demand fluctuation pattern in the graph structure through AI, convert it into standard format data, and generate AI logistics demand prediction records.

[0023] The periodic node extraction number sequence includes node number identifier, trigger time index, period start position, and period fluctuation amplitude. The high-frequency tag index matching sequence includes dense node number, high-frequency fluctuation amplitude, window start and end time, and tag positioning weight. The periodic high-frequency composite positioning number sequence includes composite node number, node intersection index, periodic and high-frequency attribute identifier, and node time sequence mark. The graph path link structure number table includes path number, node connection relationship, node interval distance, and graph mapping index. The AI ​​logistics demand prediction record includes prediction time node, demand change trend, and standard structure data format.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S111: Obtain the sequence of consecutive order quantities of materials, perform difference calculation on the order quantities of two adjacent orders in the sequence, generate a difference sequence according to the original order time order of the difference, perform sequential aggregation operation on every five consecutive difference data in the difference sequence, and generate a continuous difference sliding set. After obtaining the sequence of consecutive order quantities for materials, the difference between the order quantities of every two orders is calculated. Specifically, let the first order be... The number of orders is , No. The number of orders is The difference is calculated as follows: Taking ten orders with quantities [100, 110, 115, 120, 90, 85, 130, 125, 140, 135] as an example, the difference between the first two orders is calculated as follows: The second and third strokes are By analogy, the difference sequence [10, 5, 5, -30, -5, 45, -5, 15, -5] is obtained. Then, the difference sequence is used to construct a sliding set in chronological order. For example, the first group takes the first five differences [10, 5, 5, -30, -5], the second group takes [5, 5, -30, -5, 45], and so on. The five consecutive difference data in each group are stored in a matrix and the index is recorded. A continuous difference sliding set is constructed for subsequent calculations. In actual procurement, this sliding window method can effectively monitor the quantity changes between consecutive orders and generate the required sliding set.

[0025] S112: Based on each group of difference data in the continuous difference sliding set, and perform summation on the absolute value of all difference data, if the judgment condition of the summation value being greater than the periodic quantity fluctuation judgment threshold is met for the first time, the time index position of the corresponding difference group is recorded, and the fluctuation threshold trigger index value is generated. For each set of difference data in the continuous difference sliding set, perform absolute value summation. Let the th... Group difference is The sum of their absolute values ​​is For example, the sum of the first group [10, 5, 5, -30, -5] is... The second group [5, 5, -30, -5, 45] sums up to The summation value is compared with a threshold for determining periodic quantity fluctuations. This threshold can be set based on actual business experience, for example, 70. Record the time index position of this group. Here, the sum of the second group is 90 > 70, so the recorded index position is the trigger point, and the fluctuation threshold trigger index value is obtained. Table 1 lists the sum of the absolute values ​​of the differences in each group and whether they exceed the threshold. As shown in Table 1, when the summation value is greater than the set threshold, the trigger node is marked and the index position is recorded.

[0026] S113: Based on the fluctuation threshold, the index value is used to locate the corresponding time point in the difference sequence. Taking the time point as the starting point of the periodic node, all subsequent index positions that meet the same judgment conditions are extracted in sequence. All extracted index values ​​are arranged in chronological order to generate a periodic node extraction number sequence. Based on the fluctuation threshold, the index value is used to locate the time point in the difference sequence. For example, if the index position is recorded in the 2nd and 3rd groups, the time point corresponding to the 2nd group is used as the starting point of the periodic node. All subsequent index positions that meet the same threshold condition are retrieved and arranged in chronological order, such as extracting indexes [2, 3, 6]. These indexes are numbered to form a sequence, and finally, a periodic node extraction number sequence is generated. This number sequence can be directly used to mark and track periodic fluctuation nodes in actual material management.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S211: Extract the number sequence based on the periodic nodes, obtain the material order quantity records within the corresponding time period, and construct the order quantity subsequences with equal length intervals before and after the nodes on the time axis based on the index of each node. Set a sliding window structure of the same length for each subsequence, extract and record the three order quantity values ​​in each window one by one according to the time order, and generate the sliding window order difference unit. Based on the time index value extracted from the number sequence of the periodic node, the corresponding data frame of material order quantity records is retrieved. Each number point is used as the center position of the sliding window construction. Each sliding window covers one order data before and after that number time point, forming an order quantity sequence of length 3. Data aggregation is performed on each order sequence in chronological order, and a continuous sliding data window is generated using the time index as the primary key. Let the periodic node index be... For index 5, construct a sliding window interval as follows: The corresponding order quantities are as follows: The number of orders in the window is recorded in a three-column structure and numbered according to the construction order to form a structured data set. In actual system applications, the number of orders can be taken from the order management database or the number of orders submitted per hour in the scheduling system. The sliding window can be constructed as an overlapping window structure to ensure the continuity in time and the completeness of the difference trend judgment. Based on the above aggregation operation, the sliding window structure is constructed sequentially for all periodic node time periods to obtain the sliding window order difference unit.

[0028] S212: Based on the three order quantity data of each window in the sliding window order difference unit, perform continuous difference calculations in sequence and accumulate the results. If the accumulated difference is greater than the local high-frequency fluctuation identification threshold, the corresponding window is recorded as a demand-intensive area window, and a demand fluctuation exceeding limit index set is generated. Based on the number of three orders corresponding to each window in the sliding window order difference unit, let it be set as follows: The differences between consecutive orders within the window are respectively and The total difference within the window is calculated based on the difference plus the legal value. That is, only the difference between the quantity of the first and third orders is calculated as the volatility indicator. In practice, for , can be obtained The threshold for identifying local high-frequency fluctuations is set to 15. If the difference D is greater than this threshold, the window is identified as a demand-intensive window. The threshold of 15 is based on the 90th percentile of historical order fluctuation statistics. After analyzing three months of order volume data, it was found that the quantity change of most three consecutive orders did not exceed 15 units. Therefore, 15 was chosen as the critical value. The experimental data is shown below: As shown in Table 2, by comparing with the threshold to determine whether it belongs to the demand-intensive window, the index value of the excess window is stored in the list to obtain the demand fluctuation excess index set.

[0029] S213: Filter windows in the time series according to the index position in the demand fluctuation excess index set, arrange all window indexes that meet the conditions in order, add high-frequency demand tags to each identified window, integrate all marked window indexes and establish a complete index sequence, and generate a high-frequency tag index matching sequence. Based on the window index numbers recorded in the demand fluctuation excess index set, a filtering operation is performed on all sliding windows in the time series, retaining only the window numbers marked in the set, and adding a label field "HF" (representing high frequency) to the corresponding windows. The original index and label value of the marked windows are combined to form a mapping group, constructing a label index structure. Finally, all window indexes with added high-frequency labels are arranged and merged in chronological order to form a one-dimensional vector structure sequence. In the actual order system, by matching the index with the label, the scheduling parameters or reserved storage configuration corresponding to the demand-intensive interval can be quickly retrieved. For example, if the index number is... The windows are marked, and the labels are written into the corresponding sequences. And construct the bi-column mapping sequence as follows: index: ; Label: The structure is output as a high-frequency tag index matching sequence.

[0030] Please see Figure 4 The specific steps of S3 are as follows: S311: Based on the periodic node, extract the number sequence and the high-frequency label index matching sequence, perform a comparison operation on the time index in the sequence, if both sequences include a certain time index, record the corresponding index as the intersection node index, and generate a periodic label intersection index set; Based on the periodic node extraction of the number sequence and the high-frequency tag index matching sequence, all time index data fields within each sequence are extracted, and a time axis reference list is established. The two index sequences are then compared item by item in chronological order. Specifically, let the periodic number sequence index set be... The high-frequency tag index sequence is Using each item in set C as the retrieval benchmark, the system searches set H for a completely identical time index value. When an equality is satisfied, the index value is recorded as the intersection node index. In the implementation scenario, a list of periodic extraction node indices and a list of high-frequency response window numbers can be exported from the order system scheduling module as input sources. During the comparison process, a dual-cursor synchronous pointer mode is used to improve index matching efficiency. After traversal, all successfully matched intersection index values ​​are recorded. For example, in this case, the matching result is... The data is written into an independent intersection array structure as input parameters for subsequent processing, generating a set of periodic label intersection indices.

[0031] S312: Based on the set of intersection indexes of periodic labels, screen the position of each index value in the original order sequence, exclude nodes that exceed the current scheduling cycle boundary or are in a data missing state, and retain the corresponding index number for the remaining valid nodes to obtain a list of available intersection node indexes; Based on the index numbers recorded in the intersection index set of periodic labels, the start and end time boundaries in the main time series of order scheduling are called, and the intersection node indexes are verified and screened sequentially. Let the system scheduling period be... indexed by intersection For example, determine whether all items fall within the scheduling period. If there is an index less than t0 or greater than t0... The item is then removed from the collection. Further checks are performed to determine if any records are missing in the order data source. If the data corresponding to an index contains null values ​​or is inaccessible, that item is also removed. The remaining items are verified valid node numbers. In practice, the status of order record fields at each index position in the order database is retrieved, and the status value field is used to indicate its validity. If the status is NULL or an invalidation flag value such as -1, it is marked as an unavailable node. Finally, the intersection node number index that passes the validity screening is retained. Number the available nodes and create a list of available intersection nodes.

[0032] S313: Based on all index numbers in the available intersection node index list, according to the order of arrangement on the order data timeline, perform number reordering, and assign a continuous number identifier to each node to generate a periodic high-frequency composite positioning number sequence; Based on the index numbers in the list of available intersection nodes, a time-based sorting operation is performed based on the main timeline structure of the order data. Ascending order is used to sort the indexes according to their actual chronological order. For example, the indexes in the list are... Then the sorted result is After sorting, the indices are assigned numbers and labels sequentially according to their positions. Let the starting number be 1; then index 4 corresponds to label 1, index 9 corresponds to label 2, and so on. After forming a mapping table structure, the indices and numbers are combined into pairs to output the number sequence structure. For example, the output format is... In the order processing system, this numbering structure can be used to locate composite response areas. By binding continuous numbering labels to the index, structured encoding is completed, and a periodic high-frequency composite positioning number sequence is established.

[0033] Please see Figure 5 The specific steps of S4 are as follows: S411: Based on the periodic high-frequency composite positioning number sequence, perform ascending sorting operation on all node indices in the sequence, rearrange each node according to the chronological order, construct the time index difference sequence between adjacent nodes, extract and number all the differences, and obtain the node interval index distance sequence. Obtain the periodic high-frequency composite location number sequence, extract the node indices of all nodes in the sequence, and perform an ascending sort operation according to time order to obtain an ordered node set. Using this ordered set as input, calculate the index interval for each pair of adjacent nodes, assuming the index interval is . Perform difference operations on all adjacent indices and record the results, combining this with the index sequence in the actual scenario. The index difference sequence can be obtained as follows: For each difference, number the node position, record the corresponding starting index and difference value to build a structured mapping table, forming a node interval information structure. The index difference can reflect the tightness of periodic high-frequency composite nodes in the time dimension. For example, an interval of more than 10 indicates that there is a sparse distribution behavior between the nodes. Then, all the interval information is collected to generate a node interval index distance sequence.

[0034] S412: Based on the node interval index distance sequence, construct a connection path structure by using each node index and its subsequent node interval index as edge weights, and sequentially connect composite nodes in the form of adjacency pairs to construct the path graph, using the formula: ; The total node connection strength is calculated and then the structure is rearranged to obtain the node path connection strength data; among which, This represents the total node connection strength. Indicates the first Node index Indicates the index of its subsequent connected nodes. This indicates the number of orders loaded between the current indexes; Based on all valid node pairs in the node interval index distance sequence, construct the connection path structure for each node pair, forming directed edges between the current node and its subsequent nodes. Assume the weight of each edge is determined by a function of the corresponding index interval and the order load item. Let the edge numbering pairs be... The number of orders loaded is The total node connection strength is obtained by formula calculation. For example, suppose the ordered node index sequence is... The corresponding number of orders loaded is Substituting into the calculation, we get: ; This result will be compared with a preset connection strength threshold, which is set to 4. This threshold is based on the fact that when the index interval between nodes exceeds 6 and the number of orders loaded is less than 5, path sparsity increases significantly, affecting graph coherence. and Given the combination of conditions, an empirical value is derived and rounded down to 4 as the judgment standard. This value increases rapidly with the increase of time interval and decrease of order density, thus becoming a key identification indicator. It is used to determine whether there are path segments with both sparse connections and uneven task load. After processing all node connection pairs, a node connection path is established, and the node path connection strength is obtained.

[0035] formula The operational logic lies in using the time interval between adjacent composite node indices. The number of orders loaded during this period Normalization is performed to obtain a measure of the "sparseness" of connections between nodes. This structure reflects a non-linear inverse relationship between time span and load density. The denominator uses... This is to compress the impact of intervals under heavy load conditions, avoid nonlinear distortion caused by direct division of nodes under heavy load, and prevent division by zero error when the order quantity is 0. Therefore, 1 is added for correction. The absolute value sign is used to avoid directional offset caused by index order, and to ensure that the connection strength is only related to the span. Finally, the cumulative connection strength of the overall graph path is formed by summing the results of all connection segments, which is used to comprehensively evaluate the overall density of the path and the rationality of scheduling.

[0036] The total node connection strength is a numerical indicator used to measure the tightness of connections between periodic high-frequency composite nodes in the graph path structure. Its calculation results comprehensively reflect the coupling relationship between the time interval between adjacent nodes and the order load within the interval. The larger the index is, the larger the index span between nodes or the smaller the order load carried between paths, indicating that the connection between two periodic nodes on the time axis is sparse and the path distribution structure is loose. Conversely, it indicates that the nodes are concentrated in time and the order processing is dense, and the path has high continuity. Therefore, this indicator can serve as an important basis for assessing composite path density and resource scheduling pressure. It is essentially a measure of aggregation strength, used to describe the structural relationship of node paths under the dual dimensions of time and task intensity.

[0037] S413: Based on the connection pair sequence structure corresponding to the node path connection strength data, establish a path graph index mapping, extract the start and end node indexes and connection numbers of each connection segment, and store all path segments in a structured order to obtain the graph path link structure number table. Based on the start and end node pairs indices of each connection segment in the node path connection strength quantity, a connection path mapping structure is constructed, and each path segment is structured and encoded, with the starting point set as... The endpoint is Connector segments are numbered sequentially starting from 1. After the structured path segments are encoded, they form connection segment mapping pairs, such as... , ,in To assign index values, a complete number table is generated for all connection paths. The table contains three columns: starting index, ending index, and path number, and is stored in ascending order by the starting index. The output structure is shown below: As shown in Table 3, each pair of composite nodes constitutes a graph path, and a mapping relationship is established with its structure number. The completed path index structure forms a graph path link structure number table.

[0038] Please see Figure 6 The specific steps of S5 are as follows: S511: Based on the map path link structure number table, obtain the corresponding structure of node index and path number, extract the time index of the start and end nodes of each path segment, reorder all path numbers in time order, construct a path number sequence structure that progresses in time, perform continuity verification on the path number sequence and mark time features, and establish a map path time sequence arrangement matrix. Based on the path link structure numbering table in the graph, the starting node index and ending node index corresponding to each path segment are extracted. An ordered node pair structure is constructed by combining the absolute index values ​​of each node on the time axis according to the path number order. All path segments are sorted according to the time index values, and the sorted path number sequence is recorded. All path numbers are mapped onto the time axis to form a sequence distribution vector. Then, the time difference between adjacent path segments is calculated to determine whether there is overlap or skipping, and a path continuity identifier vector is constructed. If the path number is... And its corresponding node time index is The path after sorting by time is: Establish the path number sequence as follows Combined with time index difference, such as This indicates stable progress between paths; conversely, if the difference in a certain segment is too large, it is determined to be a skipped path. All path segments are sorted to generate a matrix structure and the path-time mapping is recorded to generate a graph path time sequence arrangement matrix.

[0039] S512: Based on the path sequence and time index information in the time-series arrangement matrix of the map path, monitor the fluctuation pattern of the path structure in the time dimension, perform periodic behavior recognition and extract feature label information according to the path change frequency and structural jump behavior, and generate a logistics path fluctuation feature vector. Based on the sorted path structure information in the temporal arrangement matrix of the path graph, the number difference, time interval difference, and path number step count between path segments are extracted. These indicators are merged to construct a path change vector matrix to determine the path change trend. For example, when the number difference is consistently 1 while the time interval is consistently unequal, it is marked as unstable continuous fluctuation; if the path number interval is 1 while the time interval approaches a constant value, it is marked as a stable sequence. Furthermore, structural features are marked on these path behaviors. For example, in a certain segment, the path number is... The time interval is This can be categorized as a periodic pattern; conversely, if the numbering jumps to a certain value... , interval is If the result is a non-periodic mutation type, then a similarity judgment is performed on the entire path sequence and fields such as fluctuation category number, start and end index, duration length, and structure identifier are recorded in the feature label. These fields are then organized into a structured vector structure to generate a logistics path fluctuation feature vector.

[0040] S513: Based on the structural label information extracted from the logistics path fluctuation feature vector, each label is uniformly encoded and mapped, decomposed into a set of data fields in a standard format, construct a structured data frame, and establish an AI logistics demand prediction record. Based on the feature label fields recorded in the logistics path fluctuation feature vector, the label content is extracted and standardized item by item. Information such as fluctuation category identifier, sequence type, path change span, and duration are uniformly mapped to structured coding items. For example, the fluctuation category "periodic" is mapped to C1, "mutation" to C2, and the duration field, such as four time units, is standardized to "T004". The above coding results are combined into a standard structured multi-field record. Extended fields such as original path number, start and end node index, time index interval, fluctuation sequence number, and feature location information are added to the record. This forms a data frame in the output structure, with each line representing a standard prediction record. The structure fields are as follows: As shown in Table 4, the prediction results are archived in the form of structured records, integrating information such as path timing, structure type and duration to establish an AI logistics demand prediction record.

[0041] Please see Figure 7 An AI-based logistics demand forecasting system includes: The periodic difference extraction module is used to execute S1: obtain the sequence of consecutive order quantities of materials, read the quantity of adjacent orders to calculate the quantity difference, perform cumulative absolute difference judgment and mark the trigger node in sequence, and generate a periodic node extraction number sequence; The window label positioning module is used to execute S2: extract the number sequence based on the cycle node, obtain the order quantity record of the corresponding time period for material allocation, build a continuous window by sliding it over time, and when the cumulative difference within the window is greater than the local high-frequency fluctuation identification threshold, the window is marked as a demand-intensive node window and a high-frequency label index matching sequence is generated. The node intersection identification module is used to perform S3: extract the number sequence from the periodic node and match the high-frequency label index sequence, select the intersection node that has both periodic attributes and high-frequency labels, and generate a periodic high-frequency composite positioning number sequence; The graph path construction module is used to execute S4: based on the periodic high-frequency composite positioning number sequence, sort the node sequence on the time axis, record the interval index length of adjacent nodes, connect each composite node to construct the graph path, establish a graph node arrangement mapping index, and generate a graph path link structure number table. The prediction data generation module is used to execute S5: based on the graph path link structure number table, it reads the node time sequence and path number sequence, identifies the logistics demand fluctuation pattern in the graph structure through AI, and converts it into standard format data to generate AI logistics demand prediction records.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An AI-based method for predicting logistics demand, characterized in that, Includes the following steps: S1: Obtain the sequence of consecutive order quantities of materials, read the quantity of adjacent orders to calculate the quantity difference, perform cumulative absolute difference judgment and mark the trigger node in sequence, and generate a sequence of periodic node extraction numbers; S2: Extract the number sequence based on the cycle node, obtain the order quantity record of the corresponding time period for material allocation, construct a continuous window by sliding it over time, and when the accumulated difference in the window is greater than the local high-frequency fluctuation identification threshold, mark the window as a demand-intensive node window and generate a high-frequency label index matching sequence. S3: Based on the number sequence extracted from the periodic nodes and the matching sequence of the high-frequency tag index, select the intersection node that has both periodic attributes and high-frequency tags to generate a periodic high-frequency composite positioning number sequence; S4: Based on the periodic high-frequency composite positioning number sequence, sort the node sequence on the time axis, record the interval index length of adjacent nodes, connect each composite node to construct a graph path, establish a graph node arrangement mapping index, and generate a graph path link structure number table. S5: Based on the graph path link structure number table, read the node time sequence and path number sequence, identify the logistics demand fluctuation pattern in the graph structure through AI, convert it into standard format data, and generate AI logistics demand prediction records.

2. The AI-based logistics demand forecasting method according to claim 1, characterized in that, The periodic node extraction number sequence includes node number identifier, trigger time index, period start position, and period fluctuation amplitude. The high-frequency tag index matching sequence includes dense node number, high-frequency fluctuation amplitude, window start and end time, and tag positioning weight. The periodic high-frequency composite positioning number sequence includes composite node number, node intersection index, periodic and high-frequency attribute identifier, and node time sequence mark. The graph path link structure number table includes path number, node connection relationship, node interval distance, and graph mapping index. The AI ​​logistics demand prediction record includes prediction time node, demand change trend, and standard structure data format.

3. The AI-based logistics demand forecasting method according to claim 1, characterized in that, The steps for obtaining the periodic node extraction number sequence are as follows: S111: Obtain the sequence of consecutive order quantities of materials, perform difference calculation on the order quantities of two adjacent orders in the sequence, generate a difference sequence according to the original order time order of the difference, perform sequential aggregation operation on every five consecutive difference data in the difference sequence, and generate a continuous difference sliding set. S112: Based on each group of difference data in the continuous difference sliding set, and perform summation processing on the absolute value of all difference data, if the judgment condition of the summation value being greater than the periodic quantity fluctuation judgment threshold is met for the first time, the time index position of the corresponding difference group is recorded, and the fluctuation threshold trigger index value is generated. S113: Based on the fluctuation threshold, the index value is used to locate the corresponding time point in the difference sequence. Taking the time point as the starting point of the periodic node, all subsequent index positions that meet the same judgment conditions are extracted in sequence. All extracted index values ​​are arranged in chronological order to generate a periodic node extraction number sequence.

4. The AI-based logistics demand forecasting method according to claim 1, characterized in that, The steps for obtaining the high-frequency tag index matching sequence are as follows: S211: Extract the number sequence according to the periodic node, obtain the material order quantity record within the corresponding time period, and construct the order quantity subsequence of equal length interval before and after the node on the time axis based on the index of each node. Set a sliding window structure of the same length for each subsequence, extract and record the three order quantity values ​​in each window one by one according to the time order, and generate the sliding window order difference unit. S212: Based on the three order quantity data of each window in the sliding window order difference unit, perform continuous difference calculations in sequence and accumulate the results. If the accumulated difference is greater than the local high-frequency fluctuation identification threshold, record the corresponding window as a demand-intensive area window and generate a demand fluctuation exceeding limit index set. S213: Filter windows in the time series according to the index position in the demand fluctuation excess index set, arrange all window indexes that meet the conditions in order, add high-frequency demand tags to each identified window, integrate all marked window indexes and establish a complete index sequence, and generate a high-frequency tag index matching sequence.

5. The AI-based logistics demand forecasting method according to claim 1, characterized in that, The steps for obtaining the periodic high-frequency composite positioning number sequence are as follows: S311: Based on the extracted number sequence of the periodic node and the matching sequence of the high-frequency label index, perform a comparison operation on the time index in the sequence. If both sequences include a certain time index, record the corresponding index as the intersection node index and generate a periodic label intersection index set. S312: Based on the set of periodic label intersection indexes, screen the position of each index value in the original order sequence, exclude nodes that exceed the current scheduling cycle boundary or are in a data missing state, and retain the corresponding index number for the remaining valid nodes to obtain a list of available intersection node indexes. S313: Based on all index numbers in the available intersection node index list, according to the arrangement order on the order data time axis, the numbers are reordered, and each node is assigned a continuous number identifier to generate a periodic high-frequency composite positioning number sequence.

6. The AI-based logistics demand forecasting method according to claim 5, characterized in that, The comparison operation specifically involves sequentially scanning each time index value in the period number using a step-by-step matching method, and searching for the existence of a matching index in the high-frequency label sequence.

7. The AI-based logistics demand forecasting method according to claim 1, characterized in that, The steps for obtaining the map path link structure number table are as follows: S411: Based on the periodic high-frequency composite positioning number sequence, perform an ascending sort operation on all node indices in the sequence, rearrange each node according to the chronological order, construct a time index difference sequence between adjacent nodes, extract and number all the differences, and obtain a node interval index distance sequence. S412: Based on the node interval index distance sequence, construct a connection path structure by using each node index and its subsequent node interval index as edge weights, connect the composite nodes in the form of adjacency pairs to construct a path graph, calculate and obtain the total node connection strength, and perform structural organization to obtain node path connection strength data. S413: Based on the connection pair sequence structure corresponding to the node path connection strength data, establish a path graph index mapping, extract the start and end node indexes and connection numbers of each connection segment, and store all path segments sequentially in a structured order to obtain the graph path link structure number table.

8. The AI-based logistics demand forecasting method according to claim 1, characterized in that, The steps for obtaining the AI ​​logistics demand forecast records are as follows: S511: Based on the graph path link structure number table, obtain the corresponding structure of node index and path number, extract the time index of the start and end nodes of each path segment, reorder all path numbers in time order, construct a path number sequence structure that advances in time, perform continuity verification on the path number sequence and mark time features, and establish a graph path time sequence arrangement matrix. S512: Based on the path sequence and time index information in the path time sequence arrangement matrix of the map, monitor the fluctuation pattern of the path structure in the time dimension, perform periodic behavior recognition and extract feature label information according to the path change frequency and structural jump behavior, and generate a logistics path fluctuation feature vector. S513: Based on the structural label information extracted from the logistics path fluctuation feature vector, each label is uniformly encoded and mapped, decomposed into a set of data fields in a standard format, construct a structured data frame, and establish an AI logistics demand prediction record.

9. The AI-based logistics demand forecasting method according to claim 8, characterized in that, The monitoring path structure's fluctuation pattern in the time dimension is specifically determined by aggregating the length difference, time span difference, and path number increment of consecutive paths to judge the fluctuation pattern of the path sequence.

10. An AI-based logistics demand forecasting system, characterized in that, The system is used to implement the AI-based logistics demand forecasting method according to any one of claims 1-9, and the system includes: The periodic difference extraction module is used to execute S1: obtain the sequence of consecutive order quantities of materials, read the quantity of adjacent orders to calculate the quantity difference, perform cumulative absolute difference judgment and mark the trigger node in sequence, and generate a periodic node extraction number sequence; The window label positioning module is used to execute S2: extract the number sequence according to the periodic node, obtain the order quantity record of the material allocation corresponding time period, construct a continuous window by sliding according to time, and when the cumulative difference in the window is greater than the local high frequency fluctuation identification threshold, the window is marked as a demand-intensive node window and a high frequency label index matching sequence is generated. The node intersection identification module is used to perform S3: based on the number sequence extracted from the periodic node and the matching sequence of the high-frequency tag index, select the intersection node that has both periodic attributes and high-frequency tags, and generate a periodic high-frequency composite positioning number sequence; The graph path construction module is used to execute S4: based on the periodic high-frequency composite positioning number sequence, sort the node sequence on the time axis, record the interval index length of adjacent nodes, connect each composite node to construct the graph path, establish a graph node arrangement mapping index, and generate a graph path link structure number table. The prediction data generation module is used to execute S5: based on the number table of the graph path link structure, read the node time sequence and path number sequence, identify the logistics demand fluctuation pattern in the graph structure through AI, and convert it into standard format data to generate AI logistics demand prediction records.

Images